Understanding Generative AI and Mastering the Art of Prompting Final quiz

Final Assessment: Generative AI and Prompting

Pass mark: 75%. You can retry as many times as you need.

1 What is the best definition of generative AI? Single answer A system that creates new content such as text, images or code. An improved search engine that retrieves existing pages. A program that only follows rules written by hand. A database of fully verified answers.

2 What is a token? Single answer A password that grants access to the tool. A syntax error in the prompt. A small piece of text handled by the model.

3 Why can a language model produce an answer that is wrong yet convincing? (several answers) Multiple answers possible It optimizes for plausible text, not for verified truth. It deliberately tries to deceive the user. Fluent, confident wording is not a sign of accuracy. Precise details such as figures and references are among the easiest to get wrong.

4 You are choosing between two assistants for a task involving internal company documents. Which factor should come first? Single answer The one with the most popular name. The one that handles your data in line with your organization's rules. The one that answers the fastest. The one with the most colorful interface.

5 Which prompt is most likely to give a useful result? Single answer "Write something about our new schedule." "Make it better." "Write an email to staff announcing the new Monday start time. Friendly tone, under 120 words, and do not invent any details beyond the ones I give you." "Be an expert and write the perfect email."

6 Which statements about examples in a prompt (few-shot) are correct? (several answers) Multiple answers possible They help the model reproduce a format and tone. Flawed examples are harmless because the model ignores mistakes. Varied, correct examples work better than near-identical ones.

7 A long conversation has become inconsistent, and the assistant ignores a rule you gave at the start. What is the most effective fix? Single answer Keep adding instructions to the same thread until it works. Switch to a different topic in the same conversation. Assume the tool is broken and stop using it. Summarize the decisions and constraints, then start a new conversation with that summary.

8 You need the output to be pasted into a spreadsheet by another program. Which instruction is most appropriate? Single answer "Return a CSV with the columns name, date and amount, no commentary, and leave empty cells blank." "Write it in an elegant paragraph." "Use whatever format you think is best."

9 Which are benefits of prompt chaining and decomposition? (several answers) Multiple answers possible You can check an intermediate result, such as extracted facts, before building on it. The final answer is guaranteed to be accurate. Each step is simpler and can be reused in other tasks. You keep control of the direction of the work.

10 You summarize a long report with an assistant. Which addition to the prompt makes checking easiest? Single answer "Make the summary sound authoritative." "Use only the report, quote the supporting sentence for each point and write 'not stated' for anything missing." "Add any background knowledge that seems relevant." "Keep it as short as possible, whatever is lost."

11 Your prompt works on one test input. What is the best next step before adding it to your library? Single answer Add it immediately: it worked. Test it on several varied inputs and score the results against defined criteria. Run the same input again until you get a better answer. Rewrite the whole prompt to be longer.

12 Which are good ways to verify an important claim produced by an AI? (several answers) Multiple answers possible Look for the claim in an official or primary source Ask the same assistant "Are you sure?" and accept a yes Open any cited source and confirm that it exists and says what is claimed Recalculate key figures with a calculator or spreadsheet

13 You want to test whether a prompt about candidate profiles treats people fairly. What is a useful check? Single answer Run the prompt once and trust the answer. Change a name or gender in the input, rerun, and compare the outputs. Assume that a polite tone means there is no bias. Remove all context from the prompt.

14 Which statements about personal data are correct? (several answers) Multiple answers possible A combination of details (job, employer, city) can identify a person even without a name. Removing the name always makes data anonymous. Using only the data that is really necessary reduces risk. Health information deserves particular care.

15 Which statement best describes the EU AI Act in general terms? Single answer It bans all uses of generative AI in Europe. It applies exactly the same rules to every AI system. It is a risk-based regulation, with stricter rules for higher-risk uses and transparency duties in some cases. It only concerns the individual users of chat assistants.

16 An assistant generates an image that looks like a real person giving a testimonial for your product. What is the most responsible approach? Single answer Publish it without a label, since it looks convincing. Label it clearly as AI-generated and avoid presenting it as a genuine testimonial. Remove only the person's name and publish it.

17 Which habits help preserve your critical thinking when using AI? (several answers) Multiple answers possible Try the problem or outline yourself before asking Accept suggestions that agree with you without question Ask for weaknesses and counterarguments

18 You want a spreadsheet formula from an assistant for data that includes employee salaries. What is the best way to proceed? Single answer Paste the whole sheet with real names and salaries. Describe the columns with invented sample rows, then test the formula on cases you can verify. Share only the salary column with the names intact. Apply the formula directly to the real file without testing.

19 What does RAG (retrieval-augmented generation) add to a language model? Single answer It retrains the whole model every time you ask a question. It removes the possibility of any error. It guarantees the answer is up to date regardless of the documents. It retrieves relevant passages from your documents and gives them to the model to answer from.

20 Which are sound principles for using an AI agent that can take actions on your behalf? (several answers) Multiple answers possible Give it only the access it needs Require your approval before sending, deleting or paying Give it full access so it never needs to ask Start with small, low-risk tasks and keep logs